Muse Spark 1.2 vs Muse Spark 1.1

Muse Spark 1.2 is the new backing model for Meta's Muse Code CLI, announced 2026-08-05, and it introduces a two-tier Standard vs Contributor pricing structure that Muse Spark 1.1 did not have.

Muse Spark 1.1, the July 2026 public-preview model, positioned Meta as a serious agentic-reasoning contender. Muse Spark 1.2 refocuses the same lineage on long-horizon, multi-agent coding — the workload the new Muse Code CLI is built around.

This page compares the two on positioning, pricing, and the practical migration decisions a team already on 1.1 needs to make. Spec deltas Meta did not itemize in the 2026-08-05 launch — context window, rate limits, benchmark scores — are called out as such; verify at https://developer.meta.com/ai/products/muse-code/.

Muse Spark 1.2 vs. Muse Spark 1.1: Side-by-Side

What actually changed in Muse Spark 1.2

The two changes Meta explicitly ships in 1.2 are a Muse Code CLI paired with the model, and a Standard vs Contributor pricing split. Muse Spark 1.1 launched as a general-purpose agentic reasoning model with a single public-preview tier and no first-class coding agent alongside it.

The 1.2 positioning is narrower and more concrete: "a purpose-built coding agent optimized for long-horizon, multi-agentic workflows and transparent auditability supported by Muse Spark 1.2." That framing signals that the model was tuned for the workload Muse Code exercises — multiple coordinating workers, background reviewers, and long sessions.

Spec deltas Meta did not itemize on 2026-08-05 include context window, rate limits, output token cap, and benchmark scores. Do not assume they carry over from 1.1 verbatim; verify at https://developer.meta.com/ai/products/muse-code/ before rewriting production limits.

• New: Muse Code CLI, event-log auditability, session-continuity memory
• New: Standard vs Contributor two-tier pricing
• Carried over from 1.1: agentic reasoning lineage
• Not itemized in launch: context window, rate limits, benchmarks — verify at https://developer.meta.com/ai/products/muse-code/

Pricing: the Standard vs Contributor decision

Standard tier is the default privacy-preserving option on Muse Spark 1.2 at $1.25 input / $0.15 cached input / $4.25 output per million tokens, and Meta does not use prompts or outputs to improve products. Contributor tier is roughly 12x cheaper — $0.10 input / $0.002 cached / $0.20 output per M — but Meta uses prompts and outputs to improve their products.

Muse Spark 1.1 had no published production pricing at launch, so most teams paying anything today for 1.1 usage are on preview allowances. Moving to 1.2 is the first time cost planning becomes possible for either model.

The Contributor tier is a real lever for internal tooling, batch code migrations, and non-sensitive automation. It is the wrong choice for client code, regulated workloads, or anything you would not put in a public gist.

• Standard: $1.25 in / $4.25 out per M — no product-improvement use of data
• Contributor: $0.10 in / $0.20 out per M — data used to improve Meta products
• Cached input dramatically cheaper on both tiers
• 1.1 preview had no published production pricing

→ Compare per-token pricing for every AI model

Access, readiness, and what to change in production

Muse Code CLI installs via a single command: curl -fsSL https://dev.meta.ai/install.sh | bash, browser-authenticated. That is the only officially documented install path on 2026-08-05.

Teams already running on 1.1 through the Model API should plan a controlled cutover: keep 1.1 for the agentic-reasoning workloads that already work, and evaluate 1.2 first on the workflows Meta calls out — multi-agent coding, long sessions, and any task where an auditable event log matters.

Across the client roster we run — HOA & condo boards, TheBotScout, DroneAndDefense — the practical question for a model swap is always the same: does the new tier let us defend the cost line to finance, and does the audit trail let us defend the workflow to counsel. Muse Spark 1.2 answers the first question with published pricing and the second with Muse Code's event log.

• Muse Code install: curl -fsSL https://dev.meta.ai/install.sh | bash
• 1.2 tuned for multi-agent coding; 1.1 tuned for general agentic reasoning
• Event-log auditability is a 1.2/Muse-Code capability, not a 1.1 capability
• Rate limits and quotas not itemized — verify at https://developer.meta.com/ai/products/muse-code/

When to stay on 1.1 vs move to 1.2

Move to Muse Spark 1.2 if your primary workload is agentic coding, long-running sessions, or anything where you need an auditable event log per run. 1.2 was built for that shape of work and pairs with the Muse Code CLI that produces the audit trail.

Stay on Muse Spark 1.1 for now if you have a general agentic-reasoning workload that already works and you do not need the coding-CLI features. There is no announced deprecation of 1.1, so a controlled evaluation is fine.

• Coding, multi-agent orchestration, auditability: 1.2 + Muse Code
• General agentic reasoning already in production on 1.1: dual-run to compare
• Cost-sensitive internal tooling: consider 1.2 Contributor tier
• Regulated or client-facing workloads: 1.2 Standard tier

The Verdict

Muse Spark 1.2 is the version to build on for coding-agent work. The pairing with Muse Code, the event log, and published two-tier pricing make it the first Muse release you can defend to finance and to legal at the same time.

Muse Spark 1.1 is not obsolete — it remains the general-purpose agentic reasoning entry in Meta's lineup as of 2026-08-05. If you are getting real value from 1.1 outside of coding workflows, keep it in place while you evaluate 1.2.

Practical move: adopt 1.2 through Muse Code for any coding workflow, default to Standard tier, and only route explicit non-sensitive workloads to Contributor. Verify context window and rate limits at https://developer.meta.com/ai/products/muse-code/ before you rewrite production quotas.